| Message ID | 20230209024022.3371768-9-robert.hu@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Linear Address Masking (LAM) KVM Enabling | expand |
On Thu, Feb 09, 2023 at 10:40:21AM +0800, Robert Hoo wrote: >When in KVM emulation, calculated a LA for data access, apply LAM if >guest is at that moment LAM active, so that the following canonical check >can pass. This sounds weird. Passing the canonical checking isn't the goal. Emulating the behavior of a LAM-capable processor on memory accesses is. > >Signed-off-by: Robert Hoo <robert.hu@linux.intel.com> >--- > arch/x86/kvm/emulate.c | 6 ++++++ > arch/x86/kvm/x86.h | 13 +++++++++++++ > 2 files changed, 19 insertions(+) > >diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c >index 5cc3efa0e21c..d52037151133 100644 >--- a/arch/x86/kvm/emulate.c >+++ b/arch/x86/kvm/emulate.c >@@ -700,6 +700,12 @@ static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt, > *max_size = 0; > switch (mode) { > case X86EMUL_MODE_PROT64: >+ /* >+ * LAM applies only on data access >+ */ one-line comments look like /* Bla bla bla */ >+ if (!fetch && is_lam_active(ctxt->vcpu)) >+ la = kvm_untagged_addr(la, ctxt->vcpu); >+ > *linear = la; > va_bits = ctxt_virt_addr_bits(ctxt); > if (!__is_canonical_address(la, va_bits)) >diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h >index 7228895d4a6f..9397e9f4e061 100644 >--- a/arch/x86/kvm/x86.h >+++ b/arch/x86/kvm/x86.h >@@ -135,6 +135,19 @@ static inline int is_long_mode(struct kvm_vcpu *vcpu) > #endif > } > >+#ifdef CONFIG_X86_64 >+static inline bool is_lam_active(struct kvm_vcpu *vcpu) Drop this function because kvm_untagged_addr() already does these checks (and taking user/supervisor pointers into consideration). >+{ >+ return kvm_read_cr3(vcpu) & (X86_CR3_LAM_U48 | X86_CR3_LAM_U57) || >+ kvm_read_cr4_bits(vcpu, X86_CR4_LAM_SUP); >+} >+#else >+static inline bool is_lam_active(struct kvm_vcpu *vcpu) >+{ >+ return false; >+} >+#endif >+ > static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu) > { > int cs_db, cs_l; >-- >2.31.1 >
On Mon, 2023-02-13 at 11:53 +0800, Chao Gao wrote: > On Thu, Feb 09, 2023 at 10:40:21AM +0800, Robert Hoo wrote: > > When in KVM emulation, calculated a LA for data access, apply LAM > > if > > guest is at that moment LAM active, so that the following canonical > > check > > can pass. > > This sounds weird. Passing the canonical checking isn't the goal. > Emulating > the behavior of a LAM-capable processor on memory accesses is. > Emm, how about describe like this: In KVM emulation, apply LAM rule for linear address calculated, (data access only), i.e. clear possible meta data in LA, before doing canonical check. > > +#ifdef CONFIG_X86_64 > > +static inline bool is_lam_active(struct kvm_vcpu *vcpu) > > Drop this function because kvm_untagged_addr() already does these > checks > (and taking user/supervisor pointers into consideration). > OK
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c index 5cc3efa0e21c..d52037151133 100644 --- a/arch/x86/kvm/emulate.c +++ b/arch/x86/kvm/emulate.c @@ -700,6 +700,12 @@ static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt, *max_size = 0; switch (mode) { case X86EMUL_MODE_PROT64: + /* + * LAM applies only on data access + */ + if (!fetch && is_lam_active(ctxt->vcpu)) + la = kvm_untagged_addr(la, ctxt->vcpu); + *linear = la; va_bits = ctxt_virt_addr_bits(ctxt); if (!__is_canonical_address(la, va_bits)) diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h index 7228895d4a6f..9397e9f4e061 100644 --- a/arch/x86/kvm/x86.h +++ b/arch/x86/kvm/x86.h @@ -135,6 +135,19 @@ static inline int is_long_mode(struct kvm_vcpu *vcpu) #endif } +#ifdef CONFIG_X86_64 +static inline bool is_lam_active(struct kvm_vcpu *vcpu) +{ + return kvm_read_cr3(vcpu) & (X86_CR3_LAM_U48 | X86_CR3_LAM_U57) || + kvm_read_cr4_bits(vcpu, X86_CR4_LAM_SUP); +} +#else +static inline bool is_lam_active(struct kvm_vcpu *vcpu) +{ + return false; +} +#endif + static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu) { int cs_db, cs_l;
When in KVM emulation, calculated a LA for data access, apply LAM if guest is at that moment LAM active, so that the following canonical check can pass. Signed-off-by: Robert Hoo <robert.hu@linux.intel.com> --- arch/x86/kvm/emulate.c | 6 ++++++ arch/x86/kvm/x86.h | 13 +++++++++++++ 2 files changed, 19 insertions(+)